Multi-well Deconvolution for Well Test Interpretations

Abstract

An important problem of numerical flow simulation in the frame of an oil field development is the need to improve the quality and quantity of the input data, in particular, the data on cross-well formation properties. Well test is the most informative means of a reservoir borehole-to-borehole survey. When studying the cross-well formation properties based on the interpretation of continuous curves of pressure variations measured by telemetry sensors on down-hole pumps, the impact of adjacent wells and high noise contamination of the obtained data shall be taken into account. In an attempt to solve this issue this article studies multi-well deconvolution as a means to study all components of the pressure variations curve. The multi-well deconvolution can identify the relevant response to a change in the operating mode of a certain well and can process this response in traditional ways. The multi-well deconvolution method makes it possible to assess and account for noise impacts on the pressure curve. This approach also simplifies the curve processing, as the diagnosis of an interpreted formation model is done easier. This work proposes a new approach to building self-influence and influence functions: we propose to set them down as a sum of elementary time functions representing individual filtration modes of the formation. The well bore impact is represented exponentially; the bilinear flow – as a fourth root; the linear flow – as a second root; the radial flow – as a logarithm; the boundary effect – as a linear function. This way the self-influence and influence functions coefficients are set down linearly and Newton method can be used to define them. This approach was tested on two synthetic curves of the bottom-hole pressure obtained by modelling. The simulation and deconvolution curves of the bottom-hole pressure converged closely and this proved that the formation parameters selected for the modelling and retrieved during the self-influence and influence curves processing are close and this, consequently, proves high efficiency of the proposed approach. #COMESYSO1120.